{"slug": "microgpt", "name": "microgpt", "category": "ai", "category_name": "AI models and assistants", "description": "", "recommenders": 6, "commenters": 10, "warners": 0, "score": 6.211, "rising": -1.96, "first": "2026-02-14", "latest": "2026-07-25", "why": "I found it very helpful to look at karpathy's microgpt", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 6, "neutral": 2, "compare": 2, "warn": 0}, "spark": [0, 0, 0, 3, 0, 0, 0, 1, 2, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "I found it very helpful to look at karpathy's microgpt", "kind": "", "icon": true, "domain": null, "github_repo": "github.com/verma7/microgpt", "official_url": "https://github.com/verma7/microgpt", "pricing": "", "aliases": ["microgpt", "microGPT", "MicroGPT"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 784840992, "item_id": 49052562, "url": "https://news.ycombinator.com/item?id=49052562", "by": "theptip", "date": "2026-07-25", "stance": "recommend", "quote": "the whole point of NanoGPT is that it\u2019s distilling all of his wisdom and taste", "excerpt": "Go read Karpathy on this, his telling will be more nuanced than mine for the specific example. But the whole point of NanoGPT is that it\u2019s distilling all of his wisdom and taste into a very dense and maximally educational version that fits on a single screen. It\u2019s simply not something that has been done before in the corpus. Planet scale infra often requires some novel ideas at the architectural layer (the engineers ", "context": "", "use_case": "", "switched_from": "", "story": "Engineering management after the cost of code collapsed"}, {"id": 783109072, "item_id": 48944317, "url": "https://news.ycombinator.com/item?id=48944317", "by": "busfahrer", "date": "2026-07-17", "stance": "recommend", "quote": "I found it very helpful to look at karpathy's microgpt", "excerpt": "For anybody else trying to grok backprop, I found it very helpful to look at karpathy's microgpt, a barebones LLM in 200 lines of python. The autograd itself is just 40 lines, and reading that code and then trying to do a clean room implementation of it is a great exercise: https://gist.github.com/karpathy/8627fe009c40f57531cb18360106ce95 https://karpathy.github.io/2026/02/12/microgpt/", "context": "", "use_case": "learning backprop", "switched_from": "", "story": "What loss.backward() actually does"}, {"id": 779268064, "item_id": 48704254, "url": "https://news.ycombinator.com/item?id=48704254", "by": "asxndu", "date": "2026-06-27", "stance": "recommend", "quote": "Definitely Andrej Kapathy's 250 line code of Micro GPT.", "excerpt": "Definitely Andrej Kapathy's 250 line code of Micro GPT. I think it can fit on someone's back. (if a small enough font is used) - https://gist.github.com/karpathy/8627fe009c40f57531cb18360106ce95 - https://karpathy.github.io/2026/02/12/microgpt/", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: Smallest amount of working ML weights that can be tattooed on a body?"}, {"id": 755268848, "item_id": 47204303, "url": "https://news.ycombinator.com/item?id=47204303", "by": "growingswe", "date": "2026-02-28", "stance": "recommend", "quote": "Great stuff!", "excerpt": "Great stuff! I wrote an interactive blogpost that walks through the code and visualizes it: https://growingswe.com/blog/microgpt", "context": "", "use_case": "", "switched_from": "", "story": "Microgpt"}, {"id": 752524960, "item_id": 47032810, "url": "https://news.ycombinator.com/item?id=47032810", "by": "kengoa", "date": "2026-02-16", "stance": "recommend", "quote": "this is a lot easier to wrap my head around", "excerpt": "Amazing work! Reminded me of LLM Visualization (https://bbycroft.net/llm) except this is a lot easier to wrap my head around and that I can actually run the training loops, which makes sense given the simplicity of the original microgpt. To give a sense of what the loss value means, maybe you can add a small explainer section as a question and add this explanation from Karpathy\u2019s blog: > Over 1,000 steps the loss dec", "context": "", "use_case": "GPT training visualization", "switched_from": "", "story": "Show HN: Microgpt is a GPT you can visualize in the browser"}]}], "caveats": [], "who": [], "tags": [], "use_cases": ["learning backprop", "minimal gpt implementation", "c++ port of microgpt", "educational gpt reference", "gpt training visualization", "learning gpt internals"], "compared": [], "makers": [{"id": 756274529, "item_id": 47267158, "url": "https://news.ycombinator.com/item?id=47267158", "by": "easygenes", "date": "2026-03-05", "stance": "neutral", "quote": "optimizing microgpt for speed", "excerpt": "This is very much in line with what I found fascinating about optimizing microgpt for speed (0). Or rather, what I was able to do with it after doing so. It's so small and so fast to train, you can really dig deep into the optimization landscape. I've spent all my free time this past week digging into it. 0: https://entrpi.github.io/eemicrogpt/ (The writeup is from a few days ago, and I'm still running experiments be", "context": "", "use_case": "tiny GPT training speed", "switched_from": "", "story": "NanoGPT Slowrun: Language Modeling with Limited Data, Infinite Compute"}, {"id": 755849008, "item_id": 47240563, "url": "https://news.ycombinator.com/item?id=47240563", "by": "easygenes", "date": "2026-03-03", "stance": "neutral", "quote": "My hobby project this week (0) has been hyper-optimizing microgpt", "excerpt": "Topical. My hobby project this week (0) has been hyper-optimizing microgpt for M5's CPU cores (and comparing to MLX performance). Wonder if anything changes under the regime I've been chasing with these new chips. 0: https://entrpi.github.io/eemicrogpt/", "context": "", "use_case": "CPU performance optimization experiment", "switched_from": "", "story": "Apple Introduces MacBook Pro with All\u2011New M5 Pro and M5 Max"}], "threads": [{"url": "https://news.ycombinator.com/item?id=47202708", "title": "Microgpt", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=47026186", "title": "Show HN: Microgpt is a GPT you can visualize in the browser", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49048212", "title": "Engineering management after the cost of code collapsed", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48873239", "title": "What loss.backward() actually does", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48700586", "title": "Ask HN: Smallest amount of working ML weights that can be tattooed on a body?", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47655408", "title": "Show HN: I built a tiny LLM to demystify how language models work", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {"recommend": 3, "neutral": 2, "compare": 1}, "2026-03": {}, "2026-04": {"compare": 1}, "2026-05": {}, "2026-06": {"recommend": 1}, "2026-07": {"recommend": 2}, "2026-08": {}, "2026-09": {}, "2026-10": {}}, "quotes": [{"id": 784840992, "item_id": 49052562, "url": "https://news.ycombinator.com/item?id=49052562", "by": "theptip", "date": "2026-07-25", "stance": "recommend", "quote": "the whole point of NanoGPT is that it\u2019s distilling all of his wisdom and taste", "excerpt": "Go read Karpathy on this, his telling will be more nuanced than mine for the specific example. But the whole point of NanoGPT is that it\u2019s distilling all of his wisdom and taste into a very dense and maximally educational version that fits on a single screen. It\u2019s simply not something that has been done before in the corpus. Planet scale infra often requires some novel ideas at the architectural layer (the engineers ", "context": "", "use_case": "", "switched_from": "", "story": "Engineering management after the cost of code collapsed"}, {"id": 783109072, "item_id": 48944317, "url": "https://news.ycombinator.com/item?id=48944317", "by": "busfahrer", "date": "2026-07-17", "stance": "recommend", "quote": "I found it very helpful to look at karpathy's microgpt", "excerpt": "For anybody else trying to grok backprop, I found it very helpful to look at karpathy's microgpt, a barebones LLM in 200 lines of python. The autograd itself is just 40 lines, and reading that code and then trying to do a clean room implementation of it is a great exercise: https://gist.github.com/karpathy/8627fe009c40f57531cb18360106ce95 https://karpathy.github.io/2026/02/12/microgpt/", "context": "", "use_case": "learning backprop", "switched_from": "", "story": "What loss.backward() actually does"}, {"id": 779268064, "item_id": 48704254, "url": "https://news.ycombinator.com/item?id=48704254", "by": "asxndu", "date": "2026-06-27", "stance": "recommend", "quote": "Definitely Andrej Kapathy's 250 line code of Micro GPT.", "excerpt": "Definitely Andrej Kapathy's 250 line code of Micro GPT. I think it can fit on someone's back. (if a small enough font is used) - https://gist.github.com/karpathy/8627fe009c40f57531cb18360106ce95 - https://karpathy.github.io/2026/02/12/microgpt/", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: Smallest amount of working ML weights that can be tattooed on a body?"}, {"id": 762543808, "item_id": 47658988, "url": "https://news.ycombinator.com/item?id=47658988", "by": "fg137", "date": "2026-04-06", "stance": "compare", "quote": "How does this compare to Andrej Karpathy's microgpt", "excerpt": "How does this compare to Andrej Karpathy's microgpt (https://karpathy.github.io/2026/02/12/microgpt/) or minGPT (https://github.com/karpathy/minGPT)?", "context": "", "use_case": "Minimal GPT implementation", "switched_from": "", "story": "Show HN: I built a tiny LLM to demystify how language models work"}, {"id": 755268848, "item_id": 47204303, "url": "https://news.ycombinator.com/item?id=47204303", "by": "growingswe", "date": "2026-02-28", "stance": "recommend", "quote": "Great stuff!", "excerpt": "Great stuff! I wrote an interactive blogpost that walks through the code and visualizes it: https://growingswe.com/blog/microgpt", "context": "", "use_case": "", "switched_from": "", "story": "Microgpt"}, {"id": 755265968, "item_id": 47204123, "url": "https://news.ycombinator.com/item?id=47204123", "by": "verma7", "date": "2026-02-28", "stance": "compare", "quote": "2x the number of lines of code (~400L), 10x the speed", "excerpt": "I wrote a C++ translation of it: https://github.com/verma7/microgpt/blob/main/microgpt.cc 2x the number of lines of code (~400L), 10x the speed The hard part was figuring out how to represent the Value class in C++ (ended up using shared_ptrs).", "context": "", "use_case": "C++ port of microgpt", "switched_from": "", "story": "Microgpt"}, {"id": 753623265, "item_id": 47101454, "url": "https://news.ycombinator.com/item?id=47101454", "by": "redclaw", "date": "2026-02-21", "stance": "neutral", "quote": "Karpathy published microGPT.py on Feb 11", "excerpt": "Karpathy published microGPT.py on Feb 11 \u2013 a complete GPT in 243 lines of pure Python, no dependencies. I spent 20 minutes porting it to Rust. Result: 0.065s vs 297.7s for 1000 training steps on the same machine. A few things I found by reading his actual source carefully that our first version got wrong: 1. KV cache \u2013 Karpathy accumulates keys/values across positions (real causal attention). Our v1 processed each to", "context": "", "use_case": "Educational GPT reference", "switched_from": "", "story": "ZeroclawGPT: Karpathy's microGPT in Rust, 4580x faster"}, {"id": 752528704, "item_id": 47033044, "url": "https://news.ycombinator.com/item?id=47033044", "by": "interloxia", "date": "2026-02-16", "stance": "neutral", "quote": "The linked inspiration has a good blog post of microgpt implemented in python.", "excerpt": "The linked inspiration has a good blog post of microgpt implemented in python. https://karpathy.github.io/2026/02/12/microgpt/ It was submitted to hn a few days ago but only received a few comments. https://news.ycombinator.com/item?id=47000263", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: Microgpt is a GPT you can visualize in the browser"}, {"id": 752524960, "item_id": 47032810, "url": "https://news.ycombinator.com/item?id=47032810", "by": "kengoa", "date": "2026-02-16", "stance": "recommend", "quote": "this is a lot easier to wrap my head around", "excerpt": "Amazing work! Reminded me of LLM Visualization (https://bbycroft.net/llm) except this is a lot easier to wrap my head around and that I can actually run the training loops, which makes sense given the simplicity of the original microgpt. To give a sense of what the loss value means, maybe you can add a small explainer section as a question and add this explanation from Karpathy\u2019s blog: > Over 1,000 steps the loss dec", "context": "", "use_case": "GPT training visualization", "switched_from": "", "story": "Show HN: Microgpt is a GPT you can visualize in the browser"}, {"id": 752350224, "item_id": 47021889, "url": "https://news.ycombinator.com/item?id=47021889", "by": "az09mugen", "date": "2026-02-14", "stance": "recommend", "quote": "Will try that micro got which seems interesting.", "excerpt": "Will try that micro got which seems interesting. But what a strange way to show 243 lines of python in 5 columns : https://karpathy.ai/microgpt.html", "context": "", "use_case": "Learning GPT internals", "switched_from": "", "story": "Show HN: MicroGPT in 243 Lines \u2013 Demystifying the LLM Black Box"}]}